Abstract
The awarding of the Olympic Games to a certain city or the announcement of a city’s Olympic bid may be considered as a news shock that affects agents’ market expectations. A news shock implies potential impacts on the dynamic adjustment process that change not only the volatility but also the long-run steady-state levels of endogenous economic variables. In this study, we contribute to and extend previous researchers’ attempts to empirically test for the Olympic Games as a news shock by implementing full structural models and by matching Olympic hosts and bidders to structurally similar countries.
Economic development cannot be explained solely on the basis of exogenous fundamental shocks (Akerlof & Shiller, 2010; Cochrane, 1994). Since the pathbreaking contributions of Beaudry and Portier (2004, 2006) regarding expectation-driven business cycles, a growing body of literature has examined the anticipated shocks, the so-called news shocks, as a potential source of economic fluctuation (see, for instance, Barsky & Sims, 2011; Davis, 2007; Jaimovich & Rebelo, 2009; Schmitt-Grohé & Uribe, 2012).
News shocks do not constitute an exogenous change in current macroeconomic fundamentals. However, these shocks may affect the agents’ current market expectations. In other words, agents receive a signal today regarding economic developments tomorrow, such as higher productivity growth, and immediately adjust their contemporaneous investment, consumption, and work decisions (Feve, Matheron, & Sahuc, 2009; Jaimovich & Rebelo, 2009). Consequently, the announcement of forthcoming shocks may significantly affect the dynamic adjustment process by changing the volatility and persistence of endogenous economic variables (Barsky & Sims, 2011; Feve et al., 2009; Jaimovich & Rebelo, 2009).
In one of the earliest attempts to empirically implement the news shock theory in the field of sports economics, Brückner and Pappa (BP, 2015) analyze the economic effects of bidding for (or hosting) the Olympic Games on macroeconomic indicators such as investment, consumption, and output. These researchers indicate that the decision to apply for the games increases the output significantly 8 and 3 years before the actual event by 0.98 and 0.77 percentage points, respectively. For Olympic hosts, the researchers also find positive effects of 1.74, 2.60, and 1.41 percentage points at 3, 4, and 5 years preceding the games, respectively. The cumulative effect on output from 10 years before the games to 7 years after the games reaches approximately 15% (BP, 2015, p. 1352).
Olympic economic statistics may be heavily influenced by political considerations, and there is minimal agreement regarding the correct measurement of the size of Olympic investments. However, even using the highest investment figures available, the average Olympic investments for the Olympics from 1992 to 2012 did not exceed an unweighted 1% of national gross domestic product (GDP) per year; this figure is heavily influenced by the cases of Barcelona, Spain, in 1992 and Athens, Greece, in 2004 (Table 1). BP’s (2015) implicit investment multipliers of 15 for the Olympic Games are notably large, compared to the majority of the latest findings in fiscal policy research with multipliers in a range of 0–1 (Coenen et al., 2012).
Olympic Investment in Relation to Host Country GDP/Investment.
Note. The GDP data are real purchasing power parity (PPP) adjusted GDP in current prices. The average GDP/investment figures are calculated as the average between the value in the year of the games and the 8 years preceding the games. The Olympic investment figures are collected from various sources (Brunet, 1995; Hotchkiss, Moore, & Zobay, 2003; Kasimati & Dawson, 2009; Mayor of London, 2013; Poynter, 2006; Preuss, 2004; Tziralis, Tolis, Tatsiopoulos, & Aravossis, 2006).
In addition, the previous econometric ex post studies of Olympic Summer Games, although admittedly not based on the news shock theory, have been less favorable. Baade and Matheson (2002) examine the employment effects of the 1996 Atlanta Games. In a different estimated model, the researchers include periods leading to 1993 and test for impacts until 1997. In some specifications, the researchers discover the negative impacts of the Atlanta Olympics. In its most optimistic estimate, the study indicates a maximum of 42,500 additional jobs in the Olympic venue counties in the state of Georgia, the United States, at least 40% of which were transitory. This figure implies a 3.42% increase in local employment in Atlanta and a 0.05% increase in U.S. employment.
Examining the 1996 Atlanta Olympic Games, Hotchkiss, Moore, and Zobay (2003) check for an alternative intervention point from 1991 to 1998 and find a best fit for 1994, comparable to two yearly leads. The researchers’ data end in 2000, which allows them to examine any Olympic effects with a maximum of four lags. The researchers isolate a level shift of employment of 17.2% in Georgia counties that are affiliated with and close to the activities of the Olympic Games in Atlanta, which can be translated into approximately 293,000 additional jobs created. In a separate analysis, the researchers indicate a trend shift in the employment of 0.2 percentage point. Using the same data but simultaneously allowing for a level shift and trend shifts, Feddersen and Maennig (2013a) are unable to reject the hypothesis that the 1996 Olympics had no significant impact on the employment figures. In a sectoral analysis of the Atlanta Games using monthly data, Feddersen and Maennig (2013b) suggest a small increase of 29,000 jobs, exclusively for the Atlanta Olympic month, exclusively in Fulton County, and exclusively in a few specific sectors.
Comparing different ex ante and ex post periods with as much as 6-year leads and 12-year lags, Jasmand and Maennig (2008) do not find any systematic income or employment effects from the Olympic Games in Munich (1972). Analyzing the Olympic Games from 1960 to 2012, Rose and Spiegel (2011) suggest a permanent export boost of 39% in Olympic host countries; however, Maennig and Richter (2012) demonstrate that these empirical findings suffer from selection bias. 1 Testing the effects of the Olympic Games in Seoul in 1988, Barcelona in 1992, Sydney in 2000, and Beijing in 2008 on tourism and foreign exchange earnings with an autoregressive integrated moving average (ARIMA) model, Mitchell and Steward (2015) exclusively find negative Olympic impacts for the host countries, with the exception of a positive level shift of tourist numbers for South Korea. 2
Thus, it appears to be worthwhile to attempt to fill the gap between the findings of BP and the remainder of the relevant literature. Notably, the BP’s study is exceptional because the remainder of the literature did not control for 10 yearly leads and lags when measuring the effects of the Olympic Games. This control may be important according to the insights of the news theory: Long before the actual event and before a city is awarded the designation of Olympic host, there may be anticipation shocks. In the following study, we also allow for 10 yearly leads and lags.
BP do not follow the much richer and multivariate approach of the majority of other analyses and thus do not consider the well-established determinants of economic growth; this approach leads to a potential omitted variable bias. In the following study, we refer to the literature on economic growth, which has identified investment growth, government spending growth, fertility, life expectancy, and human capital, among others, as key determinants (Barro, 1991, 2003). 3
In addition, the BP’s analyses compare the economic performance of some of the most privileged countries in the world, including Australia, Canada, France, Germany, Japan, the United Kingdom, and the United States, that bid for the Olympic Games to those of all other countries in the world, including much less privileged countries, such as Uganda, Burundi, and Gambia. Therefore, the results may be influenced by a sample selection bias. We use propensity score matching (PSM) to identify the countries that are structurally similar to the bidding and hosting countries but are not bidders themselves.
In this study, we combine the insights of news shock theory and apply an appropriate number of leads and lags. We also combine the insights regarding growth and business cycles from empirical studies with the conventional wisdom regarding intervention studies and the need to correctly match treatment and control groups. Overall, we do not find significant economic effects regarding the Olympic Games. We find that these results are robust to the inclusion of a substantially revised and newer data set.
Empirical Strategy and Results
Similar to BP, we rely on data from the Penn World Table (PWT), version 7.0, as described by Heston, Summers, and Aten (2011), for the 1950–2009 period. We extend these data by including the standard determinants of economic growth from the World Bank (2011), which include the fertility rate, life expectancy at birth, the stock of human capital (share of tertiary schooling), the degree of international openness, a measure of political stability, and the change in terms of trade.
The baseline empirical strategy is in accordance with BP. 4 To maintain a short presentation, we restrict ourselves to the effects on GDP per capita growth. 5 Olympic bidders and hosts are denoted as 1 in the respective year and enter the equations with 10 lags and leads to capture the possible effects. In line with BP, we also include the lagged values of the GDP growth rate and of government spending as well as country-level fixed effects and a full set of year fixed effects. Table 2 summarizes our main results, where Model (1) shows the replicated results from BP.
News Shocks and Anticipation Effects of Hosting and Bidding for the Olympic Games.
Note. Standard errors in parentheses are clustered on the country level in all models. AIC = Akaike Information Criterion; cpi = Consumer Price Index; FE = Fixed Effects; PWT = Penns World Table.
**p < .05. ***p < .01.
BP include all available countries in their estimation and run ordinary least squares (OLS) regressions that give each country the same weight. 6 To resolve the potential implicit sample selection bias, an extensive strand of literature suggests propensity score matching as a reweighting technique (see, e.g., Caliendo & Kopeinig, 2008; Heckman, Ichimura, & Todd, 1997; Imbens, 2004; Rosenbaum & Rubin, 1983; Smith & Todd, 2005).
The covariates used to estimate the propensity score are required to affect the outcome variable (i.e., GDP growth) and the probability to become a bidder for the Olympic Games; these should preferably be measured before the treatment or should not vary over time (Caliendo & Kopeinig, 2008). Because we attempt to base the matching on the earliest possible year with as many available countries as possible, we encounter a trade-off between the lower data availability in the 1950s and the possibility that later outcomes may previously be influenced by participation in the Olympic Games. We select 1970 as the year and include as covariates, the 5-year lagged values of GDP, government spending, investment, consumption, and population. We match the bidding countries using 1 to 1 nearest neighbor matching and obtain a sample in which the structural differences between the hosts/bidders and the control group countries are substantially reduced (Appendix Table A1). Appendix Table A2 lists all the countries available in our analysis as well as their inclusion in the subsamples.
Model (2) in Table 2 reports the results for the matched sample of countries. The Olympic effects are slightly lower. Most notably, the variance explained by the model is doubled compared with the R 2 of 0.17 of the BP’s estimates.
Next, in accordance with a standard literature reference on economic growth (Barro, 1991, 2003), we include the lagged growth of investment, the price level, the share of tertiary schooling, the 1/life expectancy at birth, the fertility rate, the ratio of government consumption to GDP, the openness ratio, the change in terms of trade, and the polity2 score as a measure of the institutional quality (see www.systemicpeace.org/polityproject.html for details about the polity project). Model (3) in Table 2 lists the results for the full (nonmatched) sample for the years 1960–2009. 7 This specification reduces the Olympic hosting effects and the bidding effects.
Model (4) shows the results of a regression that both (a) controls for the usual determinants of economic growth and (b) restricts the sample to countries that match Olympic bidders/hosts. The combination of these two simple perturbations reduces all anticipated effects beyond significance. The variance explained by this model specification is tripled compared with BP.
Robustness
To assess the robustness of our results, we evaluated various alternative specifications and sensitivity analyses. Model (5) in Table 3 presents the news shocks and anticipation effects when—instead of propensity score matching—the control group is restricted to Organisation for Economic Cooperation and Development (OECD) countries.
Anticipation Effects of Hosting and Bidding for the Olympic Games: Robustness and Sensitivity Analysis.
Note. Standard errors in parentheses are clustered on the country level in all models. AIC = Akaike Information Criterion; cpi = Consumer Price Index; FE = Fixed Effects; PWT = Penns World Table.
**p < .05. ***p < .01.
The matching approach inevitably leads to the exclusion of certain countries not comparable to the hosts and bidders and therefore to a decline in sample size. The resulting increase in standard errors may reflect a variance–bias trade-off, that is, the coefficients are less biased at the cost of larger standard errors (Geman, Bienenstock, & Doursat, 1992).
To assess the effect of the increase in standard errors because of reduced sample size, we employ entropy balancing as an additional approach to account for the structural differences between the treatment and control group. The procedure weights the observations in the control group, such that the moments (in our case, mean and variance) are similar to the moments of the treatment group (Hainmueller, 2012; Hainmueller & Xu, 2013). Because the weights are based on a set of prespecified balance constraints, the resulting samples are balanced by design (Freier, Schumann, & Siedler, 2015). There is no loss of observations when using this technique. Model (6) in Table 3 displays the results of a specification similar to Model (3) listed in Table 2.
Although as demonstrated in Table 1, the Olympic investment and government expenditures figures are relatively low, there may be concerns that these two covariates act as channels through which bidding for the Olympic Games affects GDP growth, in which case they could not be included as covariates. To remedy such concerns, we exclude investment and government expenditures in Model (7), which is apart from that similar to Model (6).
We use the recently substantially revised PWT 8.1 data set as a further robustness check. 8 The PWT revision implies certain fundamental changes to selected data series. Appendix Figure A1 illustrates these changes. Models (8) and (9) display the results of our Barro-augmented model with entropy balancing (Model 8) and propensity matching (Model 9). Both specifications fail to provide evidence for significant anticipation effects prior to the Olympic Games.
Finally, we use a set of alternative propensity score matching estimators, including a five nearest neighbor matching, radius matching, and kernel matching. The results are presented and explained in Appendix Table A3.
None of the robustness checks suggest any significant news shock or anticipation effects of the Olympic Games.
Summary
The news shock theory implies that the news of an Olympic bid may change the agents’ plans and activities in the regional economy. Thus, economic effects may occur before the awarding of the Olympic Games to a city/nation or if a bid for the Olympic Games is not won. Because the hosting decision occurs 7 years before the games and bid plans usually begin 10 years before the games, the majority of earlier empirical studies may be biased because these studies did not allow for sufficient leads in the empirical implementation.
This article attempts to bridge the gap between a recent study (BP, 2015) that allows for 10 yearly leads and lags (and finds significant economic effects) and earlier studies with fewer leads that did not find significant effects but did allow for structural models. In this study, we include both 10 yearly leads and lags and structural models. We also compare the Olympic host nations with other matching nations. Converse to BP (2015), who find that a country’s GDP growth rate—aggregated in a 5-year window before hosting the games—can be raised by approximately 8.23 percentage points and that the cumulative effect up to 7 years after the games adds approximately 15 percentage points, we do not find significant economic effects from the Olympic Games.
We conclude that BP’s estimations are important because they indicate the possibility of news shocks and the anticipatory effects of the Olympic Games, which may come into effect early and may have been neglected by earlier studies that did not include enough leads. Nevertheless, the BP’s results suffer both from a variable selection bias (by excluding the usual exogenous variables of structural GDP models) and from a selection bias (by not using control groups that contain countries comparable to nations bidding for or hosting the Olympic Games).
On the basis of the BP’s results, policy makers might be misguided to believe that organizing the Olympic Games is one of the most efficient approaches to fiscal spending, inducing multiplier effects of incomparable size. Their results risk that policy makers (and public opinion) will feel assured by beliefs brought forward by the usual ex ante “impact studies” on the Olympic Games, promising trillions of additional GDP, hundreds of thousands of additional jobs, a self-financing of the Olympic Games (secured by multiplier effects), and so on. Although there might be positive reasons to bid for the Olympic Games, our results provide a warning that the hopes for income effects should not be part of rational motivations.
Footnotes
Appendix
News Shock and Anticipation Effects of the Olympic Games: Alternative Matching Estimators.
| Model (10) | Model (11) | Model (12) | Model (13) | Model (14) | |
|---|---|---|---|---|---|
| L.dlgdp2 | 0.115 (0.0799) | 0.0774 (0.0482) | 0.0823 (0.0615) | 0.0501 (0.0444) | 0.0415 (0.0410) |
| L.dlgov2 | 0.0158 (0.0419) | −0.0176 (0.0341) | −0.0284 (0.0224) | −0.0119 (0.0201) | −0.0105 (0.0189) |
| L.dlinv2 | −0.0107 (0.0212) | 0.0347** (0.0142) | 0.00118 (0.0162) | 0.00857 (0.0128) | 0.0130 (0.0116) |
| L.dlcpi2 | 0.0343** (0.0161) | 0.0267 (0.0242) | 0.00378 (0.0128) | 0.0105 (0.0113) | 0.00988 (0.0107) |
| Schooling | −0.0260 (0.0191) | −0.00294 (0.0207) | −0.0171 (0.0175) | −0.0130 (0.0142) | −0.0185 (0.0135) |
| 1/Life expectancy | 2.035 (3.492) | −3.093 (4.046) | 0.393 (2.943) | −1.652 (2.145) | −0.571 (1.844) |
| Fertility | −0.953 (1.061) | 0.0640 (1.565) | −1.081 (0.933) | −0.240 (0.870) | −0.444 (0.808) |
| Openness | 0.0268 (0.0260) | 0.0565 (0.0318) | 0.0106 (0.0169) | 0.0285 (0.0153) | 0.0220 (0.0144) |
| Democracy | −0.114** (0.0492) | −0.0987 (0.0831) | −0.0966** (0.0420) | −0.0909** (0.0367) | −0.0918*** (0.0349) |
| Change ToT | −3.429 (1.843) | −1.801 (4.453) | −3.908 (2.236) | −3.558 (1.944) | −2.650 (2.075) |
| Bidding Country | −0.439 (0.719) | 0.642 (0.535) | 0.343 (0.544) | 0.304 (0.536) | 0.399 (0.528) |
| F.Bidding Country | −0.241 (0.806) | 0.355 (1.011) | −0.270 (0.696) | −0.301 (0.683) | −0.390 (0.669) |
| F2.Bidding Country | −0.703 (0.882) | −0.197 (0.664) | −0.592 (0.784) | −0.629 (0.772) | −0.639 (0.743) |
| F3.Bidding Country | 0.614 (0.627) | 0.778 (0.609) | 0.616 (0.600) | 0.591 (0.512) | 0.538 (0.508) |
| F4.Bidding Country | 0.0959 (0.517) | 0.602 (0.532) | 0.0583 (0.526) | 0.0489 (0.501) | 0.0941 (0.493) |
| F5.Bidding Country | −0.748 (0.628) | −0.630 (0.719) | −0.498 (0.577) | −0.414 (0.555) | −0.444 (0.566) |
| F6.Bidding Country | 0.165 (0.955) | −0.651 (0.915) | −0.129 (0.895) | 0.0738 (0.874) | 0.116 (0.876) |
| F7.Bidding Country | 0.233 (0.526) | 0.122 (0.545) | 0.360 (0.454) | 0.319 (0.436) | 0.262 (0.443) |
| F8.Bidding Country | 0.493 (0.428) | 0.0584 (0.405) | 0.472 (0.405) | 0.581 (0.388) | 0.669 (0.390) |
| F9.Bidding Country | 0.620 (0.586) | 0.385 (0.578) | 0.716 (0.527) | 0.616 (0.479) | 0.594 (0.473) |
| F10.Bidding Country | −0.509 (0.807) | −0.538 (0.943) | −0.636 (0.764) | −0.620 (0.761) | −0.605 (0.769) |
| Hosting Country | 0.450 (0.576) | 1.156 (0.649) | 0.505 (0.513) | 0.673 (0.480) | 0.693 (0.488) |
| F.Hosting Country | 1.121 (0.756) | 1.301 (0.928) | 1.174 (0.770) | 1.287 (0.769) | 1.249 (0.770) |
| F2.Hosting Country | 1.704 (1.196) | 1.503 (0.911) | 1.530 (1.031) | 1.683 (1.069) | 1.736 (1.051) |
| F3.Hosting Country | 0.831 (0.655) | 1.310 (0.675) | 1.033 (0.673) | 1.076 (0.633) | 1.045 (0.654) |
| F4.Hosting Country | 0.747 (0.779) | 0.770 (0.732) | 0.608 (0.789) | 0.766 (0.725) | 0.839 (0.702) |
| F5.Hosting Country | 0.799 (0.628) | 0.642 (0.603) | 0.949 (0.667) | 0.960 (0.618) | 0.928 (0.631) |
| F6.Hosting Country | 0.397 (1.316) | 0.516 (1.021) | 0.405 (1.154) | 0.519 (1.175) | 0.547 (1.175) |
| F7.Hosting Country | 0.470 (0.641) | 0.756 (0.851) | 0.376 (0.604) | 0.277 (0.546) | 0.207 (0.537) |
| F8.Hosting Country | −1.703 (1.956) | −1.895 (1.921) | −1.974 (1.855) | −1.821 (1.829) | −1.795 (1.819) |
| F9.Hosting Country | 0.129 (0.520) | 0.316 (0.638) | 0.100 (0.492) | 0.185 (0.411) | 0.141 (0.414) |
| F10.Hosting Country | 0.211 (0.745) | 0.725 (0.932) | −0.0337 (0.607) | 0.0865 (0.587) | 0.0531 (0.592) |
| Lagged government expenditure and GDP | YES | YES | YES | YES | YES |
| Year FE | YES | YES | YES | YES | YES |
| Country FE | YES | YES | YES | YES | YES |
| Barro | YES | YES | YES | YES | YES |
| OECD only | — | — | — | — | — |
| PWT | 7.0 | 7.0 | 7.0 | 7.0 | 7.0 |
| Entropy balancing | — | YES | — | — | — |
| Propensity score matching | YES | — | 5NN | Radius | Kernel |
| R 2 | 0.433 | 0.505 | 0.453 | 0.409 | 0.379 |
| AIC | 6,288.2 | 13,682.8 | 8,391.7 | 13,196.1 | 13,521.5 |
| Observations | 1106 | 2414 | 1495 | 2337 | 2342 |
Note. As a reference, Model (10) corresponds to Model (4) of Table 2 and Model (11) corresponds to Model (6) of Table 3 in the main article. Instead of the default (Model (10)) matching estimator, which uses 1 to 1 nearest neighbor matching, the results of Models (12) to (14) are generated by alternative matching estimators: Model (12) matches each treatment observations to the five nearest neighbors. Model (13) uses radius matching with a caliper of 0.001. Each treatment observation is matched to all control observations within a 0.01 propensity score radius. Model (14) uses kernel matching with a biweight kernel and a bandwidth of 0.06. PWT = Penns World Table; ToT = Terms of Trade.
**p < .05. ***p < .01.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.